Golgi Cells Operate As State-Specific Temporal Filters at the Input Stage of the Cerebellar Cortex

17004 • The Journal of Neuroscience, December 15, 2010 • 30(50):17004–17014

Behavioral/Systems/Cognitive Golgi Cells Operate as State-Specific Temporal Filters at the Input Stage of the Cerebellar Cortex

Shane A. Heine,1 Stephen M. Highstein,2 and Pablo M. Blazquez1 1Department of Otolaryngolgy, Washington University, St Louis, Missouri 63110, and 2Marine Biological Laboratory, Woods Hole, Massachusetts 02543

Cerebellar processing of incoming information begins at the synapse between mossy fibers and granule cells, a synapse that is strongly controlled through Golgi cell inhibition. Thus, Golgi cells are uniquely positioned to control the flow of information into the cerebellar cortex and understanding their responses during behavior is essential to understanding cerebellar function. Here we show, for the first time, that Golgi cells express a unique oculomotor-related signal that can be used to provide state- and time-specific filtering of granule cell activity. We used newly established criteria to identify the unique electrophysiological signature of Golgi cells and recorded these neurons in the squirrel monkey ventral paraflocculus during oculomotor behaviors. We found that they carry eye movement, but not vestibular or visual, information and that this eye movement information is only expressed within a specific range of eye positions for each neuron. In addition, simultaneous recordings of Golgi cells and nearby mossy fibers revealed that Golgi cells have the opposite directional tuning of the mossy fiber(s) that likely drive their responses, and that these responses are more sluggish than their mossy fiber counterparts. Because the mossy fiber inputs appear to convey the activity of burst–tonic neurons in the brainstem, Golgi cell responses reflect a time-filtered negative image of the motor command sent to the extraocular muscles. We suggest a role for Golgi cells in the construction of forward models of movement, commonly hypothesized as a major function of the cerebellar cortex in motor control.

Introduction sible: the Golgi cell (Simpson et al., 2005; Holtzman et al., 2006; The cerebellar cortex is among the most well studied brain struc- Barmack and Yakhnitsa, 2008). Golgi cells are the main GABAergic tures in terms of its microanatomy and synaptic connections, yet interneurons influencing the input layer. They primarily receive its function is still unknown. Past studies have revealed that the inputs from mossy fibers and granule cell axons (parallel fibers) cortical circuit includes several distinct classes of excitatory and and strongly inhibit thousands of granule cells via an impressive inhibitory interneurons whose influence is distributed among axonal arborization (Eccles et al., 1964) (see Fig. 3A). Because all three layers: an input (granular) layer, an intermediate processing signals carried by the mossy fiber inputs must pass through the (molecular) layer, and an output (Purkinje cell) layer (Ramon y input stage, it is essential to understand how these signals are Cajal, 1911). The input layer is primarily occupied by granule shaped by Golgi cell inhibition. cells, which receive glutamatergic input via mossy fibers project- One potentially fruitful way to investigate the role of Golgi cell ing from other brain areas. The granule cells in turn provide the inhibition may be to consider what signal transformations are major glutamatergic input to Purkinje cells, which are the output required within the cerebellar cortex, given the signals present at neurons of the cerebellar cortex. Interspersed between this input the inputs and the outputs of the structure. A hint of the signal and output is a rich network of GABAergic interneurons that transformations performed in the cerebellar cortex is suggested influence signals at various stages in the circuit. Given all that we by a contemporary modeling approach to movement control; know about the circuit, it is astounding that the connection be- this approach suggests that the cerebellar cortex performs neural tween form and function has not yet been made. One reason for computations necessary for the construction of internal models this may be the paucity of data on the responses of the interneu- for motor control (Pasalar et al., 2006; Ebner and Pasalar, 2008). rons in alert animals, which has held back efforts to determine the For the oculomotor system, mounting evidence suggests that the processing that is performed by the cerebellar cortex during cerebellum computes an internal representation of the eye move- movement. However, definitive identification of one specific ment (forward model) from an efference copy of the motor com- class of interneuron in the alert animal has recently become fea- mand. The output of the forward model is thought to be reflected in the target neurons of ventral paraflocculus (VPFL) Purkinje cells, but not in the eye movement input neurons to the VPFL Received July 6, 2010; revised Oct. 12, 2010; accepted Oct. 18, 2010. This work was supported by National Institutes of Health Grants R01NS065099 (to S.A.H. and P.M.B.), (Ghasia et al., 2008). If this is true then the signal transformations R01EY05433 (to P.M.B. and S.M.H.), and T32GM08151 (to S.A.H.). We thank Pat Keller for help with the histology; necessary to compute a forward model of the movement would Krystal Henderson, Darryl Craig, and Valentin Militchin for technical assistance; and Dora Angelaki, Larry Snyder, need to occur within the cerebellar cortex and would most likely Angel Pastor, and Jennifer Sodini for helpful comments on the manuscript. involve interneurons such as Golgi cells. Correspondence should be addressed to Pablo M. Blazquez, 4566 Scott Avenue, St Louis, MO 63110. E-mail: We studied Golgi cells in the VPFL of the alert squirrel mon- [email protected]. DOI:10.1523/JNEUROSCI.3513-10.2010 key during a variety of vestibular and oculomotor behaviors and Copyright © 2010 the authors 0270-6474/10/3017004-11$15.00/0 provide the first evidence of a specific role of Golgi cells in filter- Heine et al. • Golgi Cells Are State-Specific Temporal Filters J. Neurosci., December 15, 2010 • 30(50):17004–17014 • 17005 ing mossy fiber granule cell throughput temporally and based on tible to large variations due to changes in firing rate because it only considers the state of the motor system. These results may have implica- adjacent interspike intervals, and is therefore better suited for analyzing non- tions for the implementation of forward models within the stationary firing rates (Holt et al., 1996). CV2s were calculated from inter- ϭ ͉ Ϫ ͉ ϩ cerebellum. spike intervals (ISIs) as CV2 2 ISIn ϩ 1 ISIn /(ISIn ϩ 1 ISIn). Preferred directions of units were calculated from spontaneous eye movements in two ways. First, for the directional specificity analysis, we Materials and Methods used the perisaccade time histogram approach with 45 or 90 degrees of Subjects and surgery. Four adult squirrel monkeys (Saimiri sciureus; three resolution (see Results, below). After generating peristimulus time his- males, one female) were used for these experiments, two of which were tograms (PSTHs), we found the directions that produced the largest trained in oculomotor tasks (083, 087) and two of which were behavior- perisaccadic increase and decrease in firing rate and considered these the ally naive (066, 078). Surgery was performed aseptically under 1–2% preferred directions for on and off responses, respectively. We then cal- isoflurane anesthesia to implant a scleral search coil for monitoring eye culated the mean change in firing rate within a 100 ms window following position and a stainless steel post for head restraint. After the monkeys the first significant change in firing rate for the preferred direction and were initially trained (two monkeys) or acclimated to head restraint (two compared this value with the mean change in firing rate during the same monkeys), a second surgery was performed to implant a stainless steel time epoch for each of the three other directions. The first significant recording chamber aimed at the cerebellar floccular complex. Additional change in firing rate was defined as the first time that the firing rate rose details of the surgical procedures have been described previously above or dropped below 2 SDs of the mean firing rate during a presaccade (Blazquez et al., 2003). Ninety percent of the neurons reported here come control period and stayed above or below for at least 100 ms. The time from the two trained monkeys, with the remaining data from untrained difference between saccade initiation (50°/s velocity threshold) and this monkeys confirming the general applicability of the results. All proce- first significant change in firing rate was taken to be the latency of the dures conformed to the National Institutes of Health Guide for the Care neuron. We analyzed on and off responses independently. Because we and Use of Laboratory Animals and were approved by the Washington desired a finer spatial resolution for the preferred direction vectors in the University Institutional Animal Care and Use Committee. paired Golgi cell–mossy fiber analysis, we used a second approach in Neural recording. Single unit and multiunit recordings were made in which we calculated a saccade vector for each saccade Ͼ2° in amplitude the cerebellar flocculus using 2–5 M⍀ tungsten electrodes (FHC). The that resulted in a change in firing rate, scaled each vector by the corre- raw signals were amplified, bandpass filtered between 100 Hz and 8 kHz sponding change in firing rate, and computed the vector average of all using eight-pole filters, and digitized at 40 kHz using a Power 1401 and such vectors. The direction of the vector average was taken as the pre- Spike2 software (CED). Spike times of single units were detected online ferred direction of the neuron. using a time–amplitude window discriminator (Bak) and were recorded Golgi cell initial changes in IFR were generally well approximated by a digitally as time stamps. Spikes were always resorted offline before anal- rising or falling exponential. However, because of the limited number of ysis using Spike2 template matching and principal component analysis data points (i.e., spikes) occurring during this initial phase, an exponen- routines. Eye position, laser position, and chair and drum positions were tial fitting was not always reliable on a saccade-by-saccade basis, so we sampled at 500 Hz using the same Power 1401. approximated time constants by measuring the time from the initial Recordings were confined to the ventral paraflocculus, which was rec- change in IFR to the time at which the neuron reached 63% of its maxi- ognized by its typically strong saccade-related hashing and was verified mal or minimal value. Time constants were measured from the IFR data by histological track reconstruction. Golgi cells were identified based on during spontaneous or visually guided saccades. For each neuron, five established criteria, as discussed in the Results section. Briefly, these cri- saccades were selected in the on or off directions and the time constant of teria include localization in the granular layer, large spike waveforms the neuron was taken as the mean of those five measured time constants. with wide spike widths that often remain isolated for Ͼ100 micrometers When time constants could not be measured in the off direction because of electrode travel, tonic and often extremely regular firing rates, and lack the IFR went to zero within a single interspike interval, the time constant of complex spikes. was assigned a value of zero (see Fig 2A for example neuron). Because Behavioral protocols. Head-restrained monkeys were comfortably this method of calculating time constants depends on precisely detecting seated in a primate chair mounted atop a vestibular table. Oculomotor the timing of changes in firing rate on a trial-by-trial basis, time constants training consisted of a standard water restriction protocol to motivate the could only be measured for the more regular Golgi cells (typically those Ͻ monkey to fixate and pursue a projected laser for liquid reward. Response with median CV2s 0.2). modality of Golgi cells (i.e., eye movement, vestibular, or visual) was Eye-position fields were measured from spontaneous eye-movement determined by having the monkey fixate or pursue a sinusoidally moving data using an algorithm that considered the correspondence of changes (0.2, 0.4, or 0.5 Hz; Ϯ 5, 8, or 10°) green laser projected on a screen 60 cm in firing rate with starting and ending eye positions of saccades. The in front of the monkey under one of three conditions: 1) smooth pursuit, motivation behind this approach is exemplified in Figure 5. When the in which the head was held stationary while the laser moved; 2) vestibulo- monkey made saccades in the on direction of the neuron, the firing rate ocular reflex (VOR) suppression (VORS), in which the laser was rotated increased as long as a portion of the eye movement was encompassed by in phase with the chair and the monkey was required to cancel its VORS; the eye-position field of the neuron. That is, saccades made into or out of, or 3) fixation during whole-field stimulation (F-WFS), in which the laser as well as within, the eye-position field result in changes in firing rate, but and head were held stationary and the monkey was required to maintain saccades made entirely outside of the field produce no change. Consid- fixation on the laser spot during movement of a patterned background. ering a series of small sequential saccades in the on direction starting on During the free-viewing condition, monkeys were turned around such one extreme of the oculomotor range and ending on the other, the neu- that they faced the experimental room instead of the screen and were ron will first start responding when a saccade endpoint crosses into the encouraged to make eye movements throughout their oculomotor range position field and it will stop responding when a saccade start point by the experimenter placing objects of interest at varied horizontal and crosses out of the position field. Thus, the area between the first saccade vertical positions relative to the monkey. endpoint producing a change in firing rate and the last saccade start point Data analysis. All data were imported to Matlab using the SON Library producing a change in firing rate defines the position field of a neuron. In and analyzed in Matlab (Mathworks) using custom written routines. Spike practice, this was measured using an interactive program that displayed times were converted to instantaneous firing rates (IFR) by taking the recip- the instantaneous firing rate and eye position traces around each saccade rocal of the interspike intervals. Multiunit activity was analyzed by rectifying and allowed the experimenter to indicate the change in instantaneous the raw extracellular waveform and smoothing it with a moving average filter firing rate following a saccade by marking the initial firing rate immedi- (10 ms window) to extract the envelope. The resulting waveform was down- ately preceding the saccade and the maximum firing rate during the sampled to match the sampling rate of the eye. postsaccade fixation period. These measured changes in firing rate were

The median CV2 was used to quantify the regularity of neurons. It is then sorted based on either the starting or ending eye position for the similar in principle to the coefficient of variation (CV), but it is less suscep- saccade (projected along the preferred direction vector of the neuron) 17006 • J. Neurosci., December 15, 2010 • 30(50):17004–17014 Heine et al. • Golgi Cells Are State-Specific Temporal Filters

Figure 1. Influence of Golgi cells on cerebellar cortex input layer processing and identification of Golgi cells. A, Schematic diagrams illustrating the position of Golgi cells within the cerebellar cortex circuit. Golgi cells receive mossy fiber inputs via their descending dendrites and soma, and parallel fiber inputs via their ascending dendrites. They strongly inhibit granule cells, which are the main glutamatergic input to Purkinje cells. Because granule cells receive inputs from mossy fibers and in turn provide an input to Golgi cells, this circuit configuration gives Golgi cells both feedforward and feedback control over granule cells. B, Location of electrolytic lesion (arrow) placed after recording a putative Golgi cell in the VPFL. The lesion is located in the granular layer (GL), identified as the dark regions in this nissl stain. The Purkinje cell layer (PCL) and molecular layer (ML) are also indicated for reference. Stimulation parameters: 15 uA cathodal current for 15 s. The locationoftheGolgicellrecordingswasconfirmedintwoadditionallesions.C,ScatterplotofmedianCV2 andmedianfiringratefor69Golgicells(black)and42Purkinjecells(gray)identifiedbased on the presence of complex spikes. The corresponding normalized density histograms are shown on the upper and right edges of the axes. spk/s, Spikes/s. D, Histograms of median interspike intervalsforalloftheGolgiandPurkinjecells(P-cell)showninCandcorrespondingspikewaveformsforasubsetof10representativeneuronsfromeachgroup.E,DistributionofGolgicelldistances from Purkinje cell layer for 21 neurons for which adequate depth measurements were taken. See also supplemental Figure S1, available at www.jneurosci.org as supplemental material. and averaged in 2° bins to generate two curves (see Fig. 5C). One curve a combination of eye- and vestibular-related responses. These represents the changes in firing rate for all saccade endpoints and the units often had short tracking distances, matched the character of other represents the changes in firing rate for all saccade start points. The the background hashing, and were difficult to maintain in isola- intersection of these two curves defines the eye-position field. tion for extended periods of recording. Therefore, these units Results were presumed to be mossy fibers (Lisberger and Fuchs, 1978b; To address the role of Golgi cells in cerebellar processing during Miles et al., 1980; Noda, 1986). The second units were usually Ͼ ␮ oculomotor behaviors, we recorded single-unit activity in the tonically active, had broader spikes ( 200 s peak to trough granular layer of the VPFL in alert-behaving squirrel monkeys time) and lower median firing rates, and isolation could often be Ͼ and used newly established criteria to identify the characteristic maintained for 10 min and sometimes up to 1 h. This second activity of Golgi cells in vivo. The VPFL granular layer was recog- type of unit was almost always heard in the background when nized based on the presence of eye movement-related hashing passing through the granular layer but was not always possible to activity, large unitary discharges and presumed mossy fiber dis- isolate. These units were determined to be Golgi cells based on charges, and the absence of complex spikes (Lisberger and Fuchs, the spike profiles and interspike interval distributions of mor- 1978b; Miles et al., 1980; Blazquez et al., 2003). In the granular phologically identified Golgi cells published by others (Vos et layer, we commonly encountered two kinds of single unit activ- al., 1999; Simpson et al., 2005; Holtzman et al., 2006; Prsa et ity. The first were units with narrow spikes (ϳ100 ␮s peak to al., 2009) and the absence of complex spike responses charac- trough time), which usually discharged with a burst/burst–tonic teristic of Purkinje cells (Thach, 1968). Figure 1B shows the eye movement-related response, a vestibular-related response, or location of an electrolytic lesion made after recording one of Heine et al. • Golgi Cells Are State-Specific Temporal Filters J. Neurosci., December 15, 2010 • 30(50):17004–17014 • 17007 these presumed Golgi cells, confirming its position in the spontaneous saccades, pursuit, VORS, and F-WFS. This Golgi granular layer. cell had a median CV2 of 0.09 and median firing rate of 38 We recorded a total of 69 putative Golgi and 40 Purkinje cells spikes/s. The neuron modulated to changes in eye position dur- in four squirrel monkeys (083, 087, 066, and 078). We quantified ing spontaneous saccades (Fig. 2A) and approximately in phase the spike patterns of the Golgi and Purkinje cells using their with changes in eye position during pursuit (Fig. 2B), but was median firing rate and median CV2. Holtzman et al. (2006) have unmodulated during VORS and F-WFS, indicating a lack of a previously shown that the median firing rate is the most useful vestibular or visual motion response. Figure 2, E–G, show the known criterion for identifying Golgi cells. We found that this average Golgi cell firing rate over at least five cycles plotted was true among our population of neurons as well. In addition, against eye position, head velocity, or visual motion (retinal slip) we found that Golgi cells exhibited a high variability in firing rate velocity during pursuit, VORS, or F-WFS, respectively. A regres- regularity, which we quantified using the CV2 metric (see Mate- sion fit to each curve reveals that the changes in eye movement rials and Methods) (Holt et al., 1996; Shin et al., 2007). Figure 1C during pursuit contribute the most to the firing-rate modulation ϭ Ϫ shows a scatter plot of median CV2 values versus median firing (slope 2.3, 0.06, and 0.04, respectively). This dominance of rates for all recorded Purkinje and presumed Golgi cells. Al- eye-movement responses was seen across the population of Golgi though Golgi cells were capable of reaching firing rates as high as cells tested (mean ratio of pursuit/VORS slopes, 19.5; mean ratio 100 Hz, 87% (n ϭ 60) of them had median firing rates Ͻ50 of pursuit/F-WFS slopes, 7.2) (Fig. 2H,I). Note the strong clus- spikes/s, whereas 93% (n ϭ 37) of Purkinje cells had median tering of data points along the ordinate axis in Figure 2, H and I, firing rates above this value. The median CV2 metric also allows indicating a lack of vestibular or visual motion responses by the separation of Purkinje and Golgi cells, but with more overlap population of Golgi cells. This exclusive coding of eye move- between distributions than the median firing rate. The majority ments was also present during behaviors that recruit the different ϭ of Golgi cells (61%, n 42) had CV2s lower than 0.2, indicating pathways in combination, such as head rotation during fixation a high regularity absent in the Purkinje cell population. We have of an earth-fixed target (VOR with target) (see supplemental Fig. included in our population of Golgi cells the three neurons with S2, available at www.jneurosci.org as supplemental material). median firing rates Ͼ80 spikes/s because these neurons met our Golgi cell modulation during the VOR-with-target task was iden- criteria for identification, although they responded differently tical to the modulation during pursuit in the absence of head during our tasks than the rest of the population and we suspect movement, which is consistent with an exclusive coding of the they may be unipolar brush cells (see supplemental materials, eye movement and indicates that Golgi cells in the VPFL respond available at www.jneurosci.org). For comparison with Figure 1 of to eye movements regardless of whether they are driven by the Holtzman and colleagues (2006), Figure 1D shows the median pursuit or vestibular system. ISI distribution of the neurons displayed in Figure 1C. Despite Similarly to many of the VPFL mossy fibers conveying eye the difference in species (monkey vs rat) and behavioral state movement signals (Miles et al., 1980), Golgi cells also responded (awake vs anesthetized), our Purkinje and Golgi cell ISI distribu- to eye movements during saccades. Therefore, we used sponta- tions are qualitatively similar to those of Holtzman and col- neous (free viewing condition; see Materials and Methods) and leagues, albeit with ours shifted toward shorter median ISIs (see visually guided saccades to more fully quantify the properties of supplemental materials and supplemental Fig. S1, available at the eye movement coding by Golgi cells. Figure 3 shows the di- www.jneurosci.org, for a more in depth comparison). We now versity of Golgi cell responses during spontaneous eye move- describe the characteristic responses of this population of Golgi ments for four different representative neurons (Fig. 3A–D) and cells in the context of vestibulo-oculomotor behaviors known to the population as a whole (Fig. 3E–F). The most common type of involve the VPFL. response seen in Golgi cells during spontaneous eye movements was a sudden decrease (Fig. 3A) and, in many cases, a complete Golgi cells in the VPFL exclusively code eye movements pause (Fig. 3B) in the firing rate following a saccade in a particular during visuo-oculomotor and vestibulo-oculomotor direction, which we refer to as the off direction (54/69; 78%). In behaviors some cases, the pause was preceded by an initial burst, but this The primate VPFL receives mossy fiber input from diverse was not always present, even for different saccades within the sources that convey vestibular, visuomotor, and eye movement same neuron. Following a pause, the Golgi cell firing rate gradu- information (Langer et al., 1985). Indeed, single Purkinje cells in ally recovered toward a tonic level. This recovery was usually cut the VPFL, the output neuron of the structure, also contain signals short by a response to the next spontaneous saccade. In some reflecting each of these pathways (Lisberger and Fuchs, 1978a; neurons, we were able to measure the time course of the recovery Miles et al., 1980; Noda and Warabi, 1987). Because, like Purkinje more fully by having the monkey make saccades to laser targets cells, the broad ascending dendritic trees of Golgi cells also re- and fixate for more extended periods; the recovery time constants ceive inputs from the parallel fiber system, which is thought to ranged from 110 to 820 ms (mean, 438 Ϯ 361 ms; n ϭ 21). convey converging information from different modalities (eye Additionally, 71% (49/69) of Golgi cells showed an increase in movement, vestibular, and visual), we hypothesized that single firing rate with changes in eye position, which we refer to as an on Golgi cells would respond to all major inputs to the flocculus in a response, that was noticeably distinct from a rebound following similar manner as Purkinje cells. We recorded Golgi cells while an off response (Fig. 3C). A subset of Golgi cells displayed rapid the monkeys performed tasks that isolate the signals of each mo- bursts for saccades in the on direction, similar to typical mossy dality, namely pursuit, VORS, and F-WFS (see Materials and fiber burst–tonic responses (n ϭ 12; 25%) (Fig. 3D), whereas the Methods). We recorded a total of 48 Golgi cells during the majority experienced more gradual increases in firing rate follow- pursuit-only task, 23 Golgi cells during the pursuit and VORS ing a saccade in the on direction (n ϭ 37; 75%). These gradual tasks, and 7 Golgi cells during all three tasks. Surprisingly, in increases in firing rate are not seen in VPFL mossy fiber popula- contrast to Purkinje cells, Golgi cells responded to eye move- tion (Lisberger and Fuchs, 1978b; Miles et al., 1980) and could ments but not to head movements or visual motion. Figure 2, reflect a low-pass filtering of eye movement signals by the Golgi A–D, presents the responses of a representative Golgi cell during cells. The burstiness of the population of Golgi cells is quantified 17008 • J. Neurosci., December 15, 2010 • 30(50):17004–17014 Heine et al. • Golgi Cells Are State-Specific Temporal Filters in Figure 3E as a burst–tonic ratio, calculated as the ratio of the maximum firing rate within the first 50 ms following response onset and the maximum firing rate between 100 and 150 ms from response onset. Ratios above 1 would indicate that the burst accounts for the dominant change in firing rate of the neuron. Con- trary to this, the majority of Golgi cells had burst–tonic ratios of Ͻ1 (median burst–tonic ratio, 0.49), indicating that the population of Golgi cells had gradual excitatory responses compared with burst– tonic mossy fibers. These responses were more gradual than the off-direction responses, with time constants often greater than the duration of the saccade (mean off- direction time constant, 61 Ϯ 62 ms; mean on-direction time constant, 124 Ϯ 107 ms; n ϭ 34; p Ͻ 0.005, Mann–Whitney U test) (Fig. 3F). Similarly to off-direction responses, following the initial on response, Golgi cell firing rates decayed down to a tonic rate. The time constants of this decay were often longer than the squirrel monkeys were capable of fixating on a laser target, so we were unable to calculate time constants representative of the population. However, others have found the value of this time constant in the macaque to be ϳ6.5 s on average for a similar population of presumed Golgi cells (Miles et al., 1980). Contrary to the differences in initial time constants, Golgi cells had similar latencies to the first significant change in firing rate (Ͼ2 SDs above or below mean presaccadic firing rate) for on and off responses, as measured during spontaneous saccades, and the responses tended to lag the eye movement (on, 36.4 Ϯ 65.3 ms; Figure2. Eyemovement-onlycodingbyGolgicells.A,ResponseduringspontaneouseyemovementsforarepresentativeGolgi off, 29.9 Ϯ 54.6 ms; mean Ϯ SD; p ϭ 0.88, cell. Traces, from top, show IFR, vertical eye position (V eye), and horizontal eye position (H eye). B–D, Response of the same Mann–Whitney U test). This suggests that neuronduringsmoothpursuit(B),VORsuppression(C),andF-WFS(D).E–G,Plotsofaveragefiringrateversuseyepositionduring a similarly timed input is responsible for pursuit (E), head velocity (vel) during VOR suppression (F), and retinal slip velocity (Ret. vel) during F-WFS with corresponding both the on and off responses of the Golgi regressionfits(black).E,Averagefiringrateduringhorizontal(black)andvertical(gray)pursuit.H,I,slopeofregressionlineduring pursuitplottedversusslopeofregressionlineduringVORsuppression(VORC;H)orF-WFS(I)forallneuronsrecordedduringthese cells but that the temporal dynamics of the ϭ ϭ input, or the Golgi cell response to the in- tasks (H, n 23; I, n 7). Laser pos, Laser position; spk, spike. See also supplemental Figure S2, available at www.jneurosci.org as supplemental material. put, is different for on and off responses. Because others have reported that Golgi Materials and Methods) (Fig. 4A). Figure 4, B and C, presents the cells in the oculomotor vermis (OMV) have broad directional tuning for saccades (Prsa et al., 2009) and that Golgi cells in crus I/II have results of this analysis for 49 neurons with significant on responses large cutaneous receptive fields (Vos et al., 1999; Holtzman et al., and 54 neurons with significant off responses. The numbers along 2006), we sought to determine how broadly tuned Golgi cells in the the abscissa indicate the cardinal directions with the highest, second VPFL are for eye movements. To address this question, we analyzed highest, third highest, and lowest response for each neuron, and the the tuning of on and off responses during spontaneous eye move- ordinate axis indicates the response magnitude for that direction, ments. To ensure that we had a sufficient number of saccades to normalized by the response in the preferred cardinal direction. Val- produce reliable averages for a large number of cells, we segmented ues of 1 for every direction would indicate that the neuron has an the oculomotor space into four 90° zones centered on each of the omnidirectional response, and values of 1 for only the preferred cardinal directions and assigned each saccade to a zone based on the direction (direction 1) would indicate a narrowly tuned neuron to direction of the saccade vector. We then computed PSTHs by bin- that cardinal direction. In support of the latter, 65% of neurons ning together Golgi cell spikes, aligned on saccade onset, for all sac- (32/49) showed an on response for the second-most responsive di- cades falling within a zone. This gave us four separate PSTHs, each rection that was Ͻ50% of the maximal on response (in the preferred representing the perisaccadic activity of the Golgi cell for saccade direction), and 96% (47/49) showed a less-than-half maximal re- directions falling within Ϯ45° of each of the cardinal directions (see sponse for the third-most responsive direction. Likewise, 70% of Heine et al. • Golgi Cells Are State-Specific Temporal Filters J. Neurosci., December 15, 2010 • 30(50):17004–17014 • 17009

for three or fewer 45° zones, and 58% of the cells had greater- than-half maximal responses for exactly three zones. This argues both that Golgi cells are narrowly tuned, and that a sizeable proportion of them have noncardinal preferred directions. Indeed, when preferred directions were approximated from the four- zone PSTHs for all Golgi cells by taking the vector average of responses to the first and second maximal directions, 46% of the neurons had preferred on directions and 51% had preferred off directions Ͼ15° from a cardinal direction. Furthermore, on- and off-directional preferences tended to be counterweighted. In 85% (34/40) of the neurons that had both significant on and off responses, the preferred directions for the on and off responses pointed in opposite directions (Fig. 4A).

Golgi cell responses have eye-position fields The Golgi cell shown in Figure 2, A–G, had an apparent saturation in firing rate during both saccades and pursuit. We determined that this saturation was not due to intrinsic properties of the neuron such as spike refractoriness, but was instead related to an eye-position threshold. Figure 5 presents this phenomenon more fully for a representative neuron. When the monkey pursued a sinusoidally moving target centered 5° to the left, the firing rate of the neuron modulated smoothly with changes in eye position (Fig. 5A, left). However, when the monkey pursued a moving target centered 5° to the right, the neuron was unmodulated (Fig. 5A, right). We call the active eye-position range of a Golgi cell the eye-position field of the neuron and we differentiate between eye-position fields for on and off responses. Note that there Figure3. TemporalpropertiesofGolgicellresponses.A–D,Representativeoff(A,B)andon is a difference in baseline firing rate between the two conditions. (C, D) responses of four different Golgi cells during saccades. Top, IFR; bottom, horizontal eye This is the result of accumulated firing rate increases for eye position (Heye). E, Distribution of burst–tonic ratios for 49 Golgi cells with significant on re- movements within the eye-position field of the neuron due to the sponses. Arrows in C and D indicate regions used to calculate burst–tonic ratios (see text). F, Distributions of on (top) and off (bottom) initial time constants for 34 Golgi cells. spk, Spike. long time constant of decay. The eye-position field of this same Golgi cell can also be seen during the spontaneous eye movements produced during the free viewing condition (Fig. 5B), in- neurons (38/54) showed an off response for the second-most re- dicating that the position fields are present during both pursuit sponsive direction that was Ͻ50% of the maximal off response, and and saccades. Note in Figure 5B that a rightward (positive) sac- 93% (50/54) showed a less-than-half maximal response for the cade starting ϳϪ7° resulted in a corresponding change in firing third-most responsive direction. Thus, our VPFL Golgi cells were rate of the neuron (black arrow), whereas a rightward saccade of more narrowly tuned than the oculomotor vermis Golgi cells of Prsa a similar amplitude starting ϳ2° had no effect on the firing rate and colleagues (2009), as a majority of their Golgi cells would be (gray arrow). By applying an algorithm that looks for changes in expected to have greater-than-half maximal responses for at least firing rate resulting from saccades with many different start and three zones under our analysis method. The narrow directional tun- end points covering the entire oculomotor range of the monkey ing of our Golgi cells was further supported by analysis of vertical and (see Materials and Methods), we determined that this Golgi cell horizontal pursuit data obtained from 29 Golgi cells (supplemental had an eye-position field between Ϫ18 and 2° for increases in Fig. S3, available at www.jneurosci.org as supplemental material). firing rate (on direction) (Fig. 5C, left). That is, for rightward (on Seventy-six percent of the neurons (n ϭ 22) modulated at least twice direction) saccades, this neuron was not responsive to saccades as much during pursuit in the preferred plane compared with the starting and ending at ϽϪ18° or Ͼ2°, but it was responsive to orthogonal plane, with a median ratio of 5.5 (slope preferred/slope saccades starting or ending within the range defined by these two orthogonal). This result stands in contrast to Golgi cell responses boundaries. Likewise, for leftward saccades (off direction) (Fig. reported in the oculomotor vermis and crus I/II, and may reveal a 5C, right), this neuron was not responsive for saccades made functional difference between the ventral paraflocculus and other outside a range of Ϫ14 to 5°. Figure 6, A and B, show the distri- cerebellar areas. bution of on eye-position fields for 19 Golgi cells and off eye- The preferred directions calculated during spontaneous sac- position fields for 20 Golgi cells for which we had sufficient data cades were uniformly distributed among the four zones for both to apply our algorithm. We confined our analysis of spontaneous ϭ Ͻ on (n 14, 11, 15, and 9; ipsilateral, contralateral, up, and down, eye movements to only those Golgi cells with median CV2s 0.2 respectively) and off responses (n ϭ 13, 11, 13, and 17; ipsilateral, because the regularity of the spike times allowed us to detect contralateral, up, and down, respectively). However, not all neu- changes in instantaneous firing rate on a saccade-by-saccade ba- rons were tuned for the cardinal directions. The fact that many sis without relying on averaging. Using this approach, we found cells did not respond strongly in the third-most responsive direc- that individual Golgi cells have eye-position fields distributed tion indicates that many of the cells were tuned for noncardinal throughout the squirrel monkey oculomotor range, with the directions. We confirmed this by generating eight-zone PSTHs population blanketing the entire range and being centered ap- for 24 cells with a sufficient number of spontaneous saccades and proximately around the center of gaze, but with individual Golgi found that 92% of cells had greater-than-half maximal responses cell fields only occupying a portion of the total range. The mean 17010 • J. Neurosci., December 15, 2010 • 30(50):17004–17014 Heine et al. • Golgi Cells Are State-Specific Temporal Filters eye position field size across the population was 15.7 Ϯ 7.4° in the on direction and 17.9 Ϯ 8.0° in the off direction, with a strong correlation between the size of the on and off fields on a per neuron basis for the 17 neurons in which we were able to calculate both the on and off position fields (Pearson correlation coefficient, 0.81; p ϽϽ 0.05) (Fig. 6C). In addition, there was a high degree of overlap between the on and off position fields for a given neuron, such that the eye position at which a Golgi cell first began to respond with a decrease in firing rate in the off direction was usually within a few degrees of the eye position at which a Golgi cell stopped responding in the on direction. Figure 6D shows a plot of the upper response field border for the off direction versus the upper response field border for the on direction for the 17 neu- Ϯ rons with sufficient data to measure both Figure4. DirectionaltuningofGolgicells.A,PSTHsofaGolgicellresponsetospontaneoussaccadeswithin 45°ofeach the on and off position fields. Most points cardinal direction. Gray dotted lines indicate 2 SDs above and below the control firing rate, which was used to calculate the first significant increase or decrease in firing rate, respectively. Bin size is 5 ms. Center plot indicates absolute depth of align along the unity line, indicating a cor- modulation for each of the four directions. Ipsi, Ipsilateral; contra, contralateral. The distance from the center of the circle respondence between these two borders to the perimeter equals 20 spikes/s (spk/s). B, C, Number of directions with on (B)oroff(C) responses. Each line in the top for most neurons. This suggests that a panel represents a single neuron and the dots indicate the normalized change in firing rate (⌬FR) for saccades in a similarly tuned input to the Golgi cells ac- particular cardinal direction zone. The directions were ranked by response amplitude such that the numbers along the counts for both the on and off responses abscissa indicate the most- to least-responsive directions, with direction 1 being the preferred direction. For both B and C, of the neurons. We address the nature of the bottom panel shows the mean and SD for all neurons. See also supplemental Figure S3, available at www.jneurosci.org this input in the next section. as supplemental material.

Golgi cell relationship with mossy fiber input As depicted in the cerebellar cortex circuit schematic in Figure 1A, Golgi cells receive glutamatergic input via two separate pathways: a direct mossy fiber input to the Golgi cell soma and descending dendrites and a feedback input via the parallel fibers (Eccles et al., 1967; Chan-Palay and Palay, 1971). The mossy fiber synapses are known to be strong (Kanichay and Silver, 2008) and the parallel fiber synapses relatively weak (Dieudonne, 1998), so we wondered to what extent the mossy fiber input could explain the Golgi cell firing-rate responses described above. In nine recording sessions, we were able to record simultaneously from a Golgi cell and either an isolated mossy fiber single unit (n ϭ 5) or multiunit hashing activity made up of one or a few single units that could not be fully isolated (n ϭ 4). The single unit activity was thought to represent mossy fiber terminals rather than granule or other cell types because the spike profile and response type matched the characteristics Figure 5. Eye position fields of a single Golgi cell. A, Response of Golgi cell during pursuit of a target to the left (left) or right previously described for mossy fibers (Lis- (right) of the center of gaze. Top, IFR; bottom, horizontal eye position (H eye). B, Response of same Golgi cell during spontaneous berger and Fuchs, 1978b; Miles et al., eye movements. Arrows indicate on response (black) or absence of on response (gray) for two saccades of similar amplitude, but 1980; Noda, 1986) and because the im- with different starting positions. C, Calculated on (left) and off (right) eye position fields for the same neuron. For both panels, the pedance of our electrodes was probably graycurveindicateschangesinfiringrateduringsaccadesversussaccadestartpointsandtheblackcurveindicateschangesinfiring too low to reliably isolate small, densely rate during saccades versus saccade end points. The shaded region is the intersection of these two curves, which defines the eye packed neurons such as granule cells. The position field of the neuron (see Materials and Methods). Heine et al. • Golgi Cells Are State-Specific Temporal Filters J. Neurosci., December 15, 2010 • 30(50):17004–17014 • 17011

Golgi cell on responses coincide with cessations of mossy fiber tonic activity, a relatively weaker stimulus. This suggests that Golgi cell time constants are a reflection of both the intrinsic membrane properties of the neurons (Forti et al., 2006) and the level of activity of the mossy fibers that innervate them. The apparent antiphasic relationship was further explored by calculating the preferred direction vectors for both the mossy fiber and Golgi cell, which were found to point in opposite directions for this pair and for the population of paired recordings as a whole (Fig. 7C). The relative latencies from saccade initiation of mossy fiber bursts and Golgi cell pauses and vice versa suggest that both the mossy fiber increases and decreases in firing rate precede the corresponding changes in Golgi cell firing rate (Fig. 7D). Finally, for the nine pairs tested, there was a strong correlation between the eye-position activation threshold for a mossy fiber and the off-direction upper eye-position field boundary of the corresponding Golgi cell (Pearson correlation coefficient, 0.91; p ϽϽ 0.05). To get a better picture of how widespread the antiphasic mossy fiber–Golgi cell responses were, we analyzed an additional 10 Golgi cells for which mossy fibers had been isolated in the same folium during the same recording session, but were not recorded simultaneously. Of these, nine Golgi cells had pauses in firing rate for saccades in the on direction of a nearby mossy fiber, suggesting that the mossy fiber may have been contributing to the pause. The remaining Golgi cell had the same on direction as a mossy fiber recorded nearby. It was not clear how the mossy fiber ter- Figure 6. Eye-position fields for the population of Golgi cells. A, B, Extent of eye-position minals themselves were distributed in terms of preferred direc- responsefieldsfor19Golgicellsintheondirection(A)and20Golgicellsintheoffdirection(B). tions. On the one hand, multiunit activity was often narrowly Circles indicate borders of position fields. Open circles correspond to estimates of borders that ϭ were not clear due to an insufficient number of saccades beyond that position. Gray, Vertical tuned for eye position (n 11/13), suggesting some response eye-movement cells; black, horizontal eye-movement cells. Positive numbers are up and ipsi- homogeneity of mossy fibers in the volume of space picked up by lateral. Top histograms indicate distributions of eye-position field extents using 1° bins and our electrodes; on the other hand, mossy fibers with different summing the bins across all neurons. C, Relationship between eye-position response field sizes directional tunings were routinely recorded within the same fo- for on and off directions in 17 Golgi cells. Diagonal line indicates equal sizes. D, Relationship lium on a single electrode track (n ϭ 7/8), often Ͻ100 microme- between the eye position at which a Golgi cell first starts to respond in the off direction (upper ters apart. It is interesting to note that on those sessions where off border) and stops responding in the on direction (upper on border) for the same 17 Golgi multiple mossy fibers were recorded near a Golgi cell, the Golgi cells. Points falling along the diagonal line indicate that the on and off upper eye position field cell pause was only explained by the directional preference of one borders are the same. of the mossy fibers (n ϭ 3). Together, these data support the hypothesis that the mossy multiunit hashing was also thought to reflect the activity of one or fiber inputs contribute to the Golgi cell’s pause in firing rate and a few mossy fiber terminals because the response type and direc- suggest that Golgi cells may only sample a subset of the available tional tuning of the hashing tended to match that of mossy fiber mossy fiber and parallel fiber activity. single units recorded nearby. The results of these nine simultaneous Golgi cell–mossy fiber recording sessions are presented in Figure 7. Figure 7A shows the Discussion raw extracellular recording trace and the instantaneous firing Golgi cells exert strong inhibitory control over granule cells, plac- rates for the sorted mossy fiber and Golgi cell of one such session. ing them in a strategic position to control the information enter- Surprisingly, given the glutamatergic nature of the mossy fiber– ing the cerebellar cortex (Eccles et al., 1967). We quantified for Golgi cell synapse, the two units appear to have an antiphasic the first time the response properties of VPFL Golgi cells in the relationship, whereby an increase in firing rate of one unit is alert monkey, revealing four primary characteristics that will help accompanied by a decrease in firing rate of the other unit. This us to better understand the role of Golgi cells in cerebellar cortical relationship was confirmed by calculating a peristimulus time processing. First, VPFL Golgi cell firing rates are predominantly histogram of Golgi cell spikes aligned with respect to either the driven by ongoing eye movements. Second, VPFL Golgi cell firing mossy fiber burst (Fig. 7B, top) or pause (Fig. 7B, bottom). A clear rates change on at least two time scales, an initial increase or dip in the Golgi cell firing rate is observed following mossy fiber decrease in firing rate with mean time constants of tens to hun- bursts and a clear rise in the Golgi cell firing rate is observed dreds of milliseconds and a longer decay or rebound in firing rate following mossy fiber pauses. Note that the Golgi cell off re- with time constants of hundreds of milliseconds to tens of sec- sponses evolved faster than the on responses (initial time con- onds. Third, VPFL Golgi cells have eye-position fields covering stants), as also seen in the population of Golgi cells (cf. Fig. 3F). A only a portion of the entire oculomotor range of the monkey. potential explanation for this phenomenon becomes clear from Fourth, VPFL Golgi cell responses have an antiphasic relation- examining the simultaneous mossy fiber and Golgi cell responses. ship with nearby mossy fibers. We discuss these properties in That is, the fast Golgi cell off responses coincide with mossy fiber more depth below and offer our hypotheses about their func- bursts, which are a rapid and strong stimulus, whereas the slower tional significance in the cerebellar cortical circuit. 17012 • J. Neurosci., December 15, 2010 • 30(50):17004–17014 Heine et al. • Golgi Cells Are State-Specific Temporal Filters

Eye-movement coding by VPFL Golgi cells The Golgi cells we recorded appeared to be exclusively driven by eye movement- related inputs and were unmodulated by vestibular or visual signals. This specificity of input is surprising not only because the large ascending dendritic fields of the Golgi cells suggest that they receive a broad convergence of diverse inputs, but also because vestibular, visual, and oculomotor signals are often already combined at the level of many VPFL projecting neurons (Mustari et al., 1988; Nakamagoe et al., 2000). The fact that VPFL Golgi cells respond specifically to the eye movement inputs suggests a highly specific connec- tivity in the input layer of the cerebellar cortex that, to our knowledge, has not been previously reported. The eye- movement inputs are likely derived from brainstem areas such as the nucleus of the prepositus hypoglossi and paramedian tract, which contain burst–tonic eye movement-related signals thought to convey an efference copy of oculomotor commands (Green et al., 2007). Within the computational framework of cerebellar cortex function in motor control, this exclusive coding of an efference copy signal suggests that Golgi cells, through their regulation of granular layer throughput, may play a critical role in the construction of internal models of the oculomotor system (e.g., forward models). Figure 7. Relationship between mossy fiber (mf) and Golgi cells (gc) simultaneously recorded during spontaneous eye move- In contrast to their burst–tonic mossy fi- ments.A,Rawtracefromextracellularrecordingofamossyfiber–Golgicellpaironthesameelectrode(top)andcorrespondingIFR ber inputs, VPFL Golgi cells have relatively for the mossy fiber (middle) and Golgi cell (bottom). An upward saccade occurred ϳ90 ms and a downward saccade occurred slow time dynamics (Fig. 3E,F), with firing- ϳ300 ms, producing a burst and then a pause in the mossy fiber firing rate. Note that the Golgi cell appears to be negatively rate rise times in the on direction often coupledtothemossyfiber.B,ThisnegativecouplingisexploredfurtherinPSTHsofthesameGolgicelltriggeredonthemossyfiber burst(top)orpause(bottom).C,Thenegativecouplingbetweenthesamemossyfiber–Golgicellpairisalsoexpressedasopposite outlasting the duration of the saccade. The directional preferences for on responses (left). This is true for the entire population of nine pairs (right). D, Relationship between off-direction responses tend to occur more Golgicelloffresponselatenciesandmossyfiberonlatencies(left),andGolgicellonresponselatenciesandmossyfiberofflatencies rapidly (on average, twice as fast as on- (right) for all nine mossy fiber–Golgi cell pairs. Mossy fiber single units are shown as black dots and multiunit hashing is shown as direction responses), with a complete pause graydots.DotsfallingabovethediagonallineindicatethatthemossyfiberrespondsbeforetheGolgicell.Theclusterofthreedots in firing often occurring within a single in- (D, right, top) correspond to Golgi cells with on latencies that fall outside the range displayed in the plot. These latencies are 114, terval of the instantaneous firing rate (Figs. 113, and 157 ms. spk, Spike; contra, contralateral; ipsi, ipsilateral. 2A,3B), but with some neurons having off- direction time constants on the order of inputs, thus giving Purkinje cells their observed phase advanced eye hundreds of milliseconds. In addition, VPFL Golgi cells have a second, longer time constant for recovery from off responses and decay signal relative to the mossy fibers (Lisberger and Fuchs, 1978a). On from on responses. These longer time constants were reported by the other hand, the long time constant could be used in computa- Miles and colleagues (1980), but the shorter initial ones were not. tions involving events occurring on longer time scales, such as learn- Together, the varied time constants give these Golgi cells the prop- ing, because the time course of the decay holds a de facto short-term erties of a bandpass filter on behaviorally relevant time scales. Spe- memory of past eye positions. cifically, the majority of VPFL Golgi cells reject high-frequency Miles and colleagues (1980) previously noted that firing rates inputs such as bursts (Fig. 3F) and low-frequency inputs such as of putative Golgi cells in the flocculus often saturate at a particu- tonic eye-position signals during steady fixations. Because Golgi cells lar eye position, usually near the center of gaze. We extend this strongly inhibit granule cells, this passband would be inverted at the finding by showing that the eye position-related saturations are level of granule cells, allowing only relatively high- or low-frequency often bounded on two sides, forming an eye-position field in mossy fiber inputs to readily pass the input stage of the cerebellar which each neuron is active. These eye-position fields do not cortex. This is consistent with previously reported observations that appear to be limited to a particular hemifield, as Miles and col- granule cells exhibit short, well timed bursts in response to stimula- leagues (1980) suggested, but rather can span both hemifields tion (Chadderton et al., 2004). Such a scheme could provide a mech- (Fig. 5). Furthermore, we show that the eye position at which a anism for granule cells to compute the time derivative of mossy fiber given Golgi cell firing rate saturates in the on direction corre- Heine et al. • Golgi Cells Are State-Specific Temporal Filters J. Neurosci., December 15, 2010 • 30(50):17004–17014 • 17013 sponds, within a few degrees, with the eye position at which the same neuron begins responding in the off direction (Fig. 6). Our simultaneous recordings of mossy fibers and Golgi cells suggest a plausible mechanism for generating the eye position fields in which a given Golgi cell’s eye-position field is determined by the activation threshold and response range of the mossy fiber(s) providing its dominant input(s). Since Golgi cells in the VPFL appear to only reflect the activity of the efference copy pathway and the relationship between the mossy fibers and Golgi cells is antiphasic, the Golgi cell activity in essence provides a negative image, with an additional temporal transformation, of the motor command being sent to the extraocular muscles. The axonal fields of individual Golgi cells appear to be mostly non- overlapping (Eccles et al., 1967), suggesting that as little as a single Golgi cell will provide the main inhibitory control over a cluster of granule cells. Consequently, individual granule cells may only “see” one eye-position field. Functionally, this arrangement could create modules of granule cells defined within a volume of space, with each module governed by, at most, a few Golgi cells and each reflecting a different state- and time-filtered signal that can be combined by downstream neurons such as Purkinje cells. It is not clear how much of these results can be generalized to other areas of the cerebellum, since we found considerably higher specificity in our population of recorded VPFL Golgi cells than was seen in OMV Golgi cells (Prsa et al., 2009). Although one would hope that Golgi cells play a similar role in the processing performed in these two areas, the differences seen between Golgi cells in VPFL and OMV may be a reflection of the different roles presumably played by these two areas in oculomotor control (Ilg and Thier, 2008). More experiments will be necessary to resolve this question.

Origin of Golgi cell responses We found that Golgi cells had highly specific responses, suggesting the possibility that a small number of inputs with similar tuning define a Golgi cell’s firing rate modulation. Additionally, we found that, consistent with earlier studies, the dominant response of Golgi cells is a pause in firing rate (Holtzman et al., 2006). It is difficult to reconcile these observations with classical descriptions of the cerebellar cortical microanatomy, wherein the dominant inputs are glutamatergic, via mossy and parallel fibers (Eccles et al., 1967; Palay and Chan-Palay, 1974). Moreover, as- suming that the units we recorded simultaneously with Golgi cells were indeed mossy fibers, it is puzzling that the pairs were antiphasic. However, there are at least two mechanisms that can be invoked to explain these phenomena (Fig. 8). First, Golgi cells have been proposed to receive inhibitory input (via GABAergic Figure8. PlausiblemechanismstoexplainGolgicellresponses.A,Left,Typicalmossyfiberburst– and glycinergic synapses) from molecular layer interneurons, in- tonic response (top) to a change in eye position (pos) resulting from a saccade (middle) cluding basket and stellate cells (D’Angelo and De Zeeuw, 2009). and hypothetical firing rate versus eye position curve (bottom). Vertical dashed line indicates Little is known about the synaptic efficacy of this inhibition, but if mossy-fiber eye-position activation threshold. Right, Typical Golgi cell response for the same eye the molecular layer interneurons are driven by inputs with a sim- movement. Note that Golgi cell off response corresponds to mossy fiber burst and on response corresponds to mossy fiber pause. This antiphasic coupling results in the Golgi cell having an inverted eye ilar tuning as the mossy fiber recorded simultaneously with the positionresponserangecomparedwiththemossyfiber(dashedline).B,C,Twopossiblemechanisms Golgi cell, they would presumably produce a Golgi cell pause in to explain antiphasic coupling of mossy fiber and Golgi cell responses based on known connections response to a mossy fiber burst. This action through a third player and synaptic properties (see text). B, Mechanism 1, Indirect mossy fiber effect over Golgi cell via could explain the antiphasic relationship between the mossy fiber inhibitory interneurons receiving similarly tuned mossy fiber-granule cell input as Golgi cell. Gluta- and Golgi cell. However, it is difficult to imagine how the tight mate (Glu) released from mossy fiber terminals activates ionotropic glutamate receptors (GluR) on correlation between the mossy fiber and Golgi cell could be main- Golgi cell and inhibitory interneuron, such as stellate cell. Stellate cell then releases inhibitory neuro- tained through a third player unless the same mossy fiber pro- transmitter, such as GABA, to generate a Golgi cell firing rate pause in response to mossy fiber burst. vides strong innervation of the molecular interneurons inhibiting GluRactivationontheGolgicellgeneratesinitialburst(cf.Fig.4A)precedingthepause.C,Mechanism the Golgi cell. Another possibility is that mossy fibers act directly 2,DirectmossyfibereffectoverGolgicellviamGluR2activationofGIRKchannels.Glutamatereleased on Golgi cells through an inhibitory mechanism mediated by frommossyfiberterminalsactivatesionotropicandmetabotropicglutamate(i.e.,mGluR2)receptors on Golgi cell. The balance between inward current through GluR and outward current through GIRK mGluR2 receptor activation of G-protein coupled inward recti- determines net response of Golgi cell. fying potassium (GIRK) channels (Watanabe and Nakanishi, 17014 • J. Neurosci., December 15, 2010 • 30(50):17004–17014 Heine et al. • Golgi Cells Are State-Specific Temporal Filters

2003). This is a similar mechanism to that of the mGluR6 recep- Ilg UJ, Thier P (2008) The neural basis of smooth pursuit eye movements in tors at the photoreceptor–bipolar cell synapse in the retina the rhesus monkey brain. Brain Cogn 68:229–240. (Snellman et al., 2008). Interestingly, some of our Golgi cells Kanichay RT, Silver RA (2008) Synaptic and cellular properties of the feedforward inhibitory circuit within the input layer of the cerebellar cortex. occasionally showed a transient (1–2 spikes) burst in firing rate J Neurosci 28:8955–8967. immediately preceding the pause, which could reflect activation Langer T, Fuchs AF, Scudder CA, Chubb MC (1985) Afferents to the floc- of excitatory ionotropic glutamate receptors such as AMPA be- culus of the cerebellum in the rhesus macaque as revealed by retrograde fore the opening of the GIRK channels. Thus, the decreases in transport of horseradish peroxidase. J Comp Neurol 235:1–25. firing rate may result from a net dominance of mGluR2 receptor- Lisberger SG, Fuchs AF (1978a) Role of primate flocculus during rapid be- mediated inhibition over AMPA and NMDA channel-mediated havioral modification of vestibuloocular reflex. I. Purkinje cell activity excitation. The long durations of the pauses found by us and during visually guided horizontal smooth-pursuit eye movements and others (Tahon et al., 2005; Holtzman et al., 2006) are supportive passive head rotation. J Neurophysiol 41:733–763. Lisberger SG, Fuchs AF (1978b) Role of primate flocculus during rapid be- of this hypothesis, since a large proportion of mGluR2 receptors havioral modification of vestibuloocular reflex. II. Mossy fiber firing pat- have nonsynaptic localizations (Lujań et al., 1997) and are thus terns during horizontal head rotation and eye movement. J Neurophysiol likely activated by glutamate spillover, which may take place in 41:764–777. the glomeruli. A small number of mossy fibers could substantially Lujań R, Roberts JD, Shigemoto R, Ohishi H, Somogyi P (1997) Differential influence the Golgi cell firing pattern, particularly if they act via plasma membrane distribution of metabotropic glutamate receptors large en marron synapses between mossy fiber rosettes and Golgi mGluR1 alpha, mGluR2 and mGluR5, relative to neurotransmitter re- cell somata (Chan-Palay and Palay, 1971) or via glutamate spill- lease sites. J Chem Neuroanat 13:219–241. over onto Golgi cell descending dendrites that participate in the Maex R, De Schutter E (1998) Synchronization of golgi and granule cell firing in a detailed network model of the cerebellar granule cell layer. glomeruli (Eccles et al., 1967). Kanichay and Silver (2008) have J Neurophysiol 80:2521–2537. already provided electrophysiological evidence that a small num- Maex R, Vos BP, De Schutter E (2000) Weak common parallel fibre syn- ber of mossy fibers can powerfully steer Golgi cell responses. In apses explain the loose synchrony observed between rat cerebellar golgi this scheme, the parallel fiber inputs may serve to synchronize the cells. J Physiol 523:175–192. spontaneous spike times of on beam Golgi cells (Maex and De Miles FA, Fuller JH, Braitman DJ, Dow BM (1980) Long-term adaptive Schutter, 1998; Maex et al., 2000) rather than contributing signif- changes in primate vestibuloocular reflex. III. Electrophysiological obser- icantly to moment-to-moment modulations in firing rate. Fur- vations in flocculus of normal monkeys. J Neurophysiol 43:1437–1476. ther investigation using pharmacological manipulation will be Mustari MJ, Fuchs AF, Wallman J (1988) Response properties of dorsolat- eral pontine units during smooth pursuit in the rhesus macaque. J Neu- necessary to better understand the origin of the antiphasic rela- rophysiol 60:664–686. tionship between Golgi cells and mossy fibers and the generation Nakamagoe K, Iwamoto Y, Yoshida K (2000) Evidence for brainstem struc- Golgi cell eye position fields. tures participating in oculomotor integration. Science 288:857–859. Noda H (1986) Mossy fibres sending retinal-slip, eye, and head velocity References signals to the flocculus of the monkey. J Physiol 379:39–60. Barmack NH, Yakhnitsa V (2008) Functions of interneurons in mouse cer- Noda H, Warabi T (1987) Responses of Purkinje cells and mossy fibres in ebellum. J Neurosci 28:1140–1152. the flocculus of the monkey during sinusoidal movements of a visual Blazquez PM, Hirata Y, Heiney SA, Green AM, Highstein SM (2003) Cere- pattern. J Physiol 387:611–628. bellar signatures of vestibulo-ocular reflex motor learning. J Neurosci Palay S, Chan-Palay V (1974) Cerebellar cortex cytology and organization. 23:9742–9751. New York: Springer. Chadderton P, Margrie TW, Haüsser M (2004) Integration of quanta in Pasalar S, Roitman AV, Durfee WK, Ebner TJ (2006) Force field effects on cerebellar granule cells during sensory processing. Nature 428:856–860. cerebellar Purkinje cell discharge with implications for internal models. Chan-Palay V, Palay SL (1971) The synapse en marron between golgi II Nat Neurosci 9:1404–1411. neurons and mossy fibers in the rat’s cerebellar cortex. Z Anat Entwick- Prsa M, Dash S, Catz N, Dicke PW, Thier P (2009) Characteristics of re- lungsgesch 133:274–287. sponses of Golgi cells and mossy fibers to eye saccades and saccadic adap- D’Angelo E, De Zeeuw CI (2009) Timing and plasticity in the cerebellum: tation recorded from the posterior vermis of the cerebellum. J Neurosci focus on the granular layer. Trends Neurosci 32:30–40. 29:250–262. Dieudonne S (1998) Submillisecond kinetics and low efficacy of parallel Ramon y Cajal S (1911) Histologie du systeme nerveux de l’home et des fibre-Golgi cell synaptic currents in the rat cerebellum. J Physiol vertebras. Paris: Maloine. 510:845–866. Shin SL, Hoebeek FE, Schonewille M, De Zeeuw CI, Aertsen A, De Schutter E Ebner TJ, Pasalar S (2008) Cerebellum predicts the future motor state. Cer- (2007) Regular patterns in cerebellar Purkinje cell simple spike trains. ebellum 7:583–588. Eccles J, Llinas R, Sasaki K (1964) Golgi cell inhibition in the cerebellar PLoS One 2:e485. cortex. Nature 204:1265–1266. Simpson JI, Hulscher HC, Sabel-Goedknegt E, Ruigrok TJ (2005) Between Eccles J, Ito M, Szenta´gothai J (1967) The cerebellum as a neuronal main and out: linking morphology and physiology of cerebellar cortical in- chine. Heidelberg: Springer. terneurons. Prog Brain Res 148:329–340. Forti L, Cesana E, Mapelli J, D’Angelo E (2006) Ionic mechanisms of auto- Snellman J, Kaur T, Shen Y, Nawy S (2008) Regulation of ON bipolar cell rhythmic firing in rat cerebellar Golgi cells. J Physiol 574:711–729. activity. Prog Retin Eye Res 27:450–463. Ghasia FF, Meng H, Angelaki DE (2008) Neural correlates of forward and Tahon K, Volny-Luraghi A, De Schutter E (2005) Temporal characteristics inverse models for eye movements: evidence from three-dimensional ki- of tactile stimuli influence the response profile of cerebellar Golgi cells. nematics. J Neurosci 28:5082–5087. Neurosci Lett 390:156–161. Green AM, Meng H, Angelaki DE (2007) A reevaluation of the inverse dy- Thach WT (1968) Discharge of Purkinje and cerebellar nuclear neurons namic model for eye movements. J Neurosci 27:1346–1355. during rapidly alternating arm movements in the monkey. J Neuro- Holt GR, Softky WR, Koch C, Douglas RJ (1996) Comparison of discharge physiol 31:785–797. variability in vitro and in vivo in cat visual cortex neurons. J Neurophysiol Vos BP, Volny-Luraghi A, De Schutter E (1999) Cerebellar Golgi cells in the 75:1806–1814. rat: receptive fields and timing of responses to facial stimulation. Eur Holtzman T, Rajapaksa T, Mostofi A, Edgley SA (2006) Different responses J Neurosci 11:2621–2634. of rat cerebellar Purkinje cells and Golgi cells evoked by widespread con- Watanabe D, Nakanishi S (2003) mGluR2 postsynaptically senses granule vergent sensory inputs. J Physiol 574:491–507. cell inputs at Golgi cell synapses. Neuron 39:821–829.